Research On Precision Optimization Algorithm For Underdetermined Blind Source Separation

Blind source separation technology is a signal processing technology that has attracted much attention and is applied to biomedical,data communication,image processing and other fields.While it is convenient for people to produce and live,it also brings some difficulties.How to separate the aliased electric wave signals in space is one of them.For many years,due to the complexity of its theoretical content and application objects,the research on blind source separation technology still has certain challenges.If you want to achieve accurate parameter estimation and subsequent processing on the target signal,you need to separate the signal effectively first.This paper will study the precision optimization of the algorithm for underdetermined blind source separation.First of all,the research significance and development status of underdetermined blind source separation are analyzed,and the classic "two-step method" for solving related problems is given.The emphasis is placed on the research situation of the mixing matrix estimation and source signal recovery in different environments.Based on the above content,the theoretical methods and system models involved in the paper are introduced in detail.Content description of sparse component analysis,linear clustering algorithm in mixing matrix estimation theory and compressed sensing theory in source signal recovery and reconstruction methods are three highlighted contents.Next,the mixing matrix estimation based on double-factor improved fuzzy C-means clustering algorithm is studied.The pre-processing and the single source point screening of the observed signal are first performed to enhance the sparsity of the signal.After the above processing,a double-factor improved fuzzy C-means clustering algorithm is proposed.By changing the membership classification of the objective function in the exponential form correction factor,and combining the fuzzy decision theory,the purpose of fast classification of data points can be achieved.The improved algorithm is applied to the separation simulation experiment of one-dimensional speech signal and analog modulation signal,and the performance of the algorithm is verified.The simulation results show that the improved algorithm can improve the accuracy of the mixing matrix estimation,and it also can change the convergence speed of the algorithm and reduce the clustering error of the objective function.Finally,the signal reconstruction based on the orthogonal matching pursuit algorithm in compressed sensing theory is studied.An algorithm combining the correlation weighted least squares dictionary learning method and the stagewise orthogonal matching pursuit algorithm is given,which can solve the F-norm minimization problem with weight signal error and change the algorithm complexity by increasing the number of atoms in a single iteration.The improved algorithm is applied to the separation simulation experiment of one-dimensional speech signal and analog modulation signal,which is divided into two situations: noiseless environment and noisy environment,and the performance of the algorithm is verified.The simulation results show that the proposed algorithm combination can improve the signal reconstruction accuracy based on the original algorithm,and it has a good compromise effect on the algorithm complexity.